Texture Analysis and Cartographic Feature Extraction.
Abstract
Investigations into using various image descriptors as well as developing interactive feature extraction software on the Digital Image Analysis Laboratory(DIAL) have culminated in a revised procedure to test statistical classification methods. An interactive experiment using this procedure was performed and showed that of the image descriptors tested, the most significant was a two component vector derived from an average and a standard deviation measure of gray shades. The texture measures failed to deliver any increase in performance for the classifier. In general, this report shows that statistical classification methods are insufficient by themselves to deliver the performance needed in a semi-automated cartographic feature extraction system. Originator-supplied keywords: Ad-Hoc image descriptor; Bayes classifier; Bhattachryya distance; Clustering; Digital Image Analysis Laboratory (DIAL); Feature extraction; Interactive processing; Laws texture measure; Principal components; Raster processing; Relaxation.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1985
- Accession Number
- ADA159220
Entities
People
- R. S. Rand
Organizations
- Geospatial Research Laboratory